current position:Home>Big Taobao technology proposed ttnet algorithm and won the award of "best industrial paper award"

Big Taobao technology proposed ttnet algorithm and won the award of "best industrial paper award"

2022-01-27 02:42:09 Heart of machine

from IEEE The computer society 、 International Network Intelligence Association 、 Sponsored by the American computer society , At the university of Oxford, 、 University of Queensland 、 deakin university 、 Chinese Academy of Medical Sciences 、 Southeast University 、 Nanjing University of Finance and Economics 、PIESAT、 Taobao technology department 、IOS Press Co sponsored second 20 the WI-IAT international conference (WI-IAT 2021) On 2021 year 12 month 14 to 17 Held in Melbourne, Australia on the th .

WI-IAT It is an annual international conference in the field of artificial intelligence , It aims to provide a broad communication platform , For academia 、 Professionals and industry people share each other's ideas and findings . WI-IAT The goal is in collective intelligence 、 Data Science 、 knowledge management 、 Network science 、 To achieve a multidisciplinary balance between autonomous agent, human centered computing and the research progress of theories and methods related to multi-agent systems . meanwhile , Committed to solving the corresponding practical application problems , It not only deepens the calculation of the future network 、 Logic 、 cognition 、 Understanding of physical and social foundations , It also makes the development and application of intelligent technology possible .

This year is WI-IAT 20 Anniversary of the .WI-IAT'21 Provides a major forum , And in all theoretical and technical fields Web Intelligence and intelligent agent technology provide high-quality services 、 Original research papers and real-world applications . The theme of this conference is 'AI in the Connected World', namely AI Connect the world , Focus on AI The application and combination of technology to practical problems , Professional researchers at home and abroad have contributed a variety of contributions AI Technology to solve practical problems . The conference invited 2 Bitmap Spirit Award winner (Leslie Vialiant Professor and Joseph Sifakis professor )、 NAE member (Tom M. Mitchell professor ) And academicians of the Royal Dutch Academy of Sciences (Frank van Harmelen professor ) Share theme reports , And the promulgation of various awards . At this meeting , Taobao technology merchant empowerment algorithm team and Zhejiang University proposed TTNet The algorithm won “ Best industrial paper award ” prize .

Internet and digital technology have opened up new sales paths for small and medium-sized enterprises , Deeply promoted the marketing mode 、 Changes in consumer operation mode and commodity production mode . However, small and medium-sized businesses are often limited by their own sales scale 、 Lack of R & D capability and market publicity , It makes it difficult for their high-quality goods to be fully explored and grow by users in a large number of goods , Therefore, how to help small and medium-sized businesses reduce costs and improve efficiency in the process of commodity growth through digitization has always been the core problem of e-commerce market mechanism optimization , It is also the basis for e-commerce platforms to help society achieve common prosperity . The enabling algorithm team of Taobao technology merchants and Zhejiang University have innovatively brought Meta-transfer Learning The migration strategy is applied to Tabular Learning On , solve Tabular Data The problem of rapid adaptation in small sample migration , It contributes a new research idea to this field , Won the general assembly “ Best industrial paper ” prize .

After a lot of research work , The team further applied this algorithm model to tap new products of high-quality and potential businesses , Build from a small amount of sparse data Tabular Data characteristics , Then apply its proposed TTNet The algorithm model accurately and efficiently predicts the new product potential of merchants , New product potential can be used to guide the large-scale support and incubation of these potential new products . Proved by experiment , The team's algorithm model can tap the potential of 100 million new products , And the effective support of tens of millions of new businesses . In this way , This technology can help the high-quality new products of small and medium-sized businesses grow rapidly on the platform , Create new profit sources for businesses , It also brings a steady stream of high-quality new products to the platform , Expand the supply side resources of the platform , Provide consumers with more diversified choices .

Tabular Data is a common data format in daily life ( Such as spreadsheets and csv file ), It mainly includes tabular data such as categories and statistical data . Most of the existing methods, such as deep neural network, often can not mine this kind of data well Tabular features , Because compared with the tree structure method, it can not well realize feature selection . secondly , The existing methods are difficult to reuse because of the complex network structure , Furthermore, it can not achieve good performance results in the migration task . Therefore, in view of these two problems , The team put forward TTNet Algorithm , Its innovation is mainly reflected in :

  1. Firstly, by designing a box table network structure , It can imitate the tree structure to realize feature selection , It can also make the neural network realize network reuse in the migration task , In order to better mine the characteristics of tabular data .
  2. Besides , The team adopted Meta-transfer To realize the rapid adaptation of migrating from source data to target data .

(3) Whether in public data sets or in practical business data applications , Compared with the existing algorithms , The team put forward TTNet All methods can achieve better performance results .

The algorithm model can be widely used in general tablular Business scenario of data , Including but not limited to new product mining in the field of e-commerce , Risk account identification in the financial field , Patient health estimates in the medical field , We are committed to solving the problem of small data and sparse samples in business scenarios , Greatly improve the accuracy and efficiency of prediction . Taobao technology merchant empowerment algorithm team and Zhejiang University applied the algorithm model in the field of e-commerce , It has greatly improved the operation ability and digital ability of small and medium-sized businesses , Help businesses expand business benefits , Contribute to the common prosperity of society . Tang Xing, vice president of new retail technology business group of Taobao business group of Alibaba group, said ,“ This is a typical paradigm for technology to really play a role in the industry , We not only bring shopping experience to consumers , Promote the digital transformation and empowerment of technology to the industry . Today's algorithm model is just a small step , Taobao technology will continue to iterate our life and Technology 、 The relationship with the world .”

copyright notice
author[Heart of machine],Please bring the original link to reprint, thank you.

Random recommended